Thesis Detección de fuego y humo en imágenes de vigilancia forestal aérea a través de redes neuronales convolucionales
Loading...
Date
2018
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Telemática
Departament
Campus
Campus Casa Central Valparaíso
Abstract
El uso de redes neuronales convolucionales es una de las sub-áreas de Deep Learning que tiene por objetivo clasificar e identificar patrones en el contexto de imágenes. Estas redes son entrenadas utilizando bases de datos de imágenes que buscan “enseñar” al modelo el comportamiento deseado, de forma tal que permita hacer mútiples tareas basadas en el reconocimiento de patrones, tales como identificar objetos, clasificar imágenes, segmentar regiones, entre otras. Este trabajo consiste en la implementación de modelos basados en redes completamente convolucionales, modelos que a diferencia de las redes convolucionales tradicionales, persiguen un procesamiento de imágenes general, entregando como resultado una imagen o tensor.El objetivo del presente trabajo es hacer detección de fuego y humo de manera rápida y efectiva, identificando forma y ubicación dentro de la imagen en prueba, bajo el contexto de monitoreo de incendios forestales. De estos modelos, el modelo propuesto (SFEwAN), es evaluado y comparado con el modelo más actual encontrado al cierre de este trabajo de título, Frizzi [16](...).
The use of Convolutional Neural Networks is an application of deep learning, which has as target to classify and identify patterns in images. These networks are trained using datasets with the aim of “teaching” the model to behave in an expected manner. This way, these models can help in pattern recognition, object identification, image classification, region segmentation, among other targets.The present work consists in an implementation of fully convolutional network models. These models are different from traditional convolutional networks because they process a general image, giving as result an image or tensor.The target of this work is to detect fire and smoke in a quick and effective fashion, identifying shape and location of object in the testing image, under the context of wildland fire monitoring. From these models, the proposed model (SFEwAN) is tested and compared to the most currently available work (at the moment this work was made), Frizzi [16](...).
The use of Convolutional Neural Networks is an application of deep learning, which has as target to classify and identify patterns in images. These networks are trained using datasets with the aim of “teaching” the model to behave in an expected manner. This way, these models can help in pattern recognition, object identification, image classification, region segmentation, among other targets.The present work consists in an implementation of fully convolutional network models. These models are different from traditional convolutional networks because they process a general image, giving as result an image or tensor.The target of this work is to detect fire and smoke in a quick and effective fashion, identifying shape and location of object in the testing image, under the context of wildland fire monitoring. From these models, the proposed model (SFEwAN) is tested and compared to the most currently available work (at the moment this work was made), Frizzi [16](...).
Description
Catalogado desde la version PDF de la tesis.
Keywords
Aprendizaje profundo, CNN, Detección de fuego, Incendios forestales, Redes neuronales
